Use FreeToken from your AI assistant

Connect your client

In a client that supports remote MCP, add a server using the endpoint below and select Streamable HTTP. Authentication is not required. Connection options vary by client; this is not a universal installation command.

Try asking

Find cataloged OpenAI-compatible models that do not require a payment card and have at least 128K context. Show the conditions, sources and verification dates, then generate a keyless config for one route.

What your assistant can query

search_models

Filter models by provider, protocol, card requirement and recorded context size.

get_provider_guide

Read official application links, eligibility conditions and existing steps.

list_price_references

Discover price reference IDs and their billing scenarios before estimating.

estimate_cost

Estimate one dated price scenario from input and output token volumes.

get_route_status

Read the latest stored route observations without running new checks.

export_keyless_config

Generate route-pack, LiteLLM or environment templates for up to five routes.

Understand the evidence

Retrieved time is not verification time. Records older than 30 days are flagged; this threshold does not certify younger records as current. Status represents sampled checks, not provider-wide uptime. Static catalog fallback is disclosed when live database reads fail.

Language coverage

These instructions and linked website pages support English, Chinese, Japanese, Spanish and Korean. Tool field names stay stable in English. v1 catalog and guide text can be English; responses disclose the content language and fallback. Set locale to choose matching website links.

{
  "name": "search_models",
  "arguments": {
    "locale": "en",
    "protocol": "OpenAI compatible",
    "accessType": "no-card",
    "minContextTokens": 131072,
    "limit": 5
  }
}

No keys, accounts or model calls

Do not send API keys, passwords, prompts or private data in requests. The service does not access your Vault, call model providers, create subscriptions or send email. Configs contain environment-variable references only. Your AI client has its own privacy policy.

Limits and troubleshooting

Requests are limited to 8 KiB; search returns at most 20 records. Anonymous traffic has an approximate limit of 60 requests per minute per IP and Cloudflare location; shared networks can share that limit. An IP is used transiently for this guard, not added to application logs. HTTP 429: wait one minute. HTTP 503: retry later; do not add a key. Browser clients are restricted to this site's origin; other clients need a server-side or native connection.

Discovery and Registry

Official Registry publication is a separate step. This page does not claim that the service is already listed, automatically enabled in your AI client or endorsed by a directory.